CN103065128A - Human body reidentification method in foggy days - Google Patents
Human body reidentification method in foggy days Download PDFInfo
- Publication number
- CN103065128A CN103065128A CN2012105927217A CN201210592721A CN103065128A CN 103065128 A CN103065128 A CN 103065128A CN 2012105927217 A CN2012105927217 A CN 2012105927217A CN 201210592721 A CN201210592721 A CN 201210592721A CN 103065128 A CN103065128 A CN 103065128A
- Authority
- CN
- China
- Prior art keywords
- human body
- vector
- image
- proper vector
- reference vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a human body reidentification method in foggy days. The method comprises that: detecting human body images in a video image; determining feature vectors of the human body images, and matching the determined feature vectors with multiple basic standard vectors in a data base collected in advance according to weight function of different colors; and using the human body images successfully matched in the data base as an identification result. Through the steps, the human body images can be confirmed in the data base, and the confirmed human body images are used as the detected human body images. Therefore, movement tracks and the activity range of each human body image in a video can be mastered especially in the foggy days. Due to the fact that the foggy days have a weakening effect on colors, human bodies are not easy to distinguish.
Description
Technical field
The present invention relates to field of video monitoring, in particular to a kind of method of identifying again at the human body in greasy weather.
Background technology
For the security protection of public place, usually adopt at present camera to realize the collection of image.
Because present video identification technology, can only identify the human body image in the video, can not confirm the corresponding individuality of human body image, thereby cause to distinguish everyone motion track, can not determine the corresponding identity of human body image in the current video.Particularly in the greasy weather, because there is the effect of reduction in the greasy weather to color, be not easy to distinguish.
Summary of the invention
The present invention aims to provide a kind of method of identifying again at the human body in greasy weather, the problem that must not confirm the individuality of human body image with solution.
In an embodiment of the present invention, provide a kind of method of identifying again at the human body in greasy weather, having comprised:
Detect the human body image in the video image;
Determine the proper vector of described human body image, according to the weight of different colours, a plurality of reference vector in the proper vector of determining and the database that gathers are in advance mated;
With the human body image that the match is successful in the described database as recognition result.
By above-mentioned step, can in database, determine human body image, with the human body image determined as the human body image that detects.Thereby can in video, grasp movement locus and the scope of activities of everyone volume image.For the situation in greasy weather, by the adding color weight, thereby improve the accuracy rate of identification.
Description of drawings
Accompanying drawing described herein is used to provide a further understanding of the present invention, consists of the application's a part, and illustrative examples of the present invention and explanation thereof are used for explaining the present invention, do not consist of improper restriction of the present invention.In the accompanying drawings:
Fig. 1 shows the process flow diagram of embodiment;
Embodiment
Below with reference to the accompanying drawings and in conjunction with the embodiments, describe the present invention in detail.Referring to Fig. 1, the step of embodiment comprises:
S11: detect the human body image in the video image;
S12: determine the proper vector of described human body image, according to the weight of different colours, a plurality of reference vector in the proper vector of determining and the database that gathers are in advance mated;
S13: with the human body image that the match is successful in the described database as recognition result.
By above-mentioned step, can in database, determine human body image, with the human body image determined as the human body image that detects.Thereby can in a plurality of human body images of video, determine unique human body image.For the situation in greasy weather, by the adding color weight, thereby improve the accuracy rate of identification.
Preferably, among the embodiment, the step of human body image comprises: use the Gaussian Background modeling to detect the moving region in video.In the moving region that detects, use based on histograms of oriented gradients (HOG) with the object detecting method of the support vector machine (latent SVM) of implicit parameter, on different scale, the human body image in the video is detected.
Preferably, among the embodiment, determine that the process of proper vector comprises:
Be the HSV form with the image transitions that detects, and extract the color distribution histogram.
Conversion from the RGB color space to the hsv color space, computing formula is as follows:
v＝max
Max=max (r, g, b) wherein, min=min (r, g, b).Such as, be the pixel of (0.1,0.2,0.5) for the RGB color value, the value in the hsv color space is (225,0.8,0.5).
Calculate color histogram:, its color is added up for each pixel in the image.For example, the v component is black less than threshold value 1, and the v component is white greater than threshold value 2 and s component less than threshold value 3, the v component between threshold value 1 and threshold value 2 and the v component be grey less than threshold value 3, other colors are colour.
For colour, evenly be divided into 6 kinds of colors according to h component from 0 to 360, namely [0,60), [60,120), [120,180), [180,240), [240,300), [300,360).
Color to each pixel is added up, and calculates every kind of color proportion in human body image, stores into successively among the array x, uses as the Characteristic of Image vector.
For example, in an image 10 pixels are arranged.Wherein black color dots and white point respectively have 3, other 4 points belong to color [60,120), this image characteristic of correspondence vector is (0.3,0.3,0,0,0.4,0,0,0,0) so.
Preferably, among the embodiment, the reference vector in the described database is determined by following steps:
Gather in advance everyone several human body images in the video image;
Adopt the K-means clustering algorithm to described several human body image computings of same person, with the one to one a plurality of proper vector computings of several human body images, obtain a proper vector of everyone correspondence as reference vector.
When using the K-means training, the color histogram of everyone volume image that obtains in the testing process is carried out cluster as proper vector, obtain the cluster centre of proper vector, and the sample that comprises of each cluster centre.
The K mean algorithm need to be inputted a parameter k, and several proper vectors.These proper vectors can be divided into k class and the sample that comprises in each class by the calculating of K mean algorithm.Like this, just the sample of input can be divided into the k class, each class represents a human body image.
The characteristic of correspondence vector that cluster centre is obtained each class stores in the database.
Above-mentioned matching process comprises:
Each proper vector corresponding to each image of computing respectively with described database in the distance of reference vector of everyone volume image;
To a plurality of distance-taxis that each proper vector obtains, determine two minimums apart from d1 and d2; Wherein, d1＜d2;
If described 1.5d1＜d2 determines that then this proper vector and the reference vector that is used for the described d1 of computing are complementary.
Determine the human body image corresponding with the nearest reference vector of each described proper vector, and everyone volume image of statistics correspondence the summation of the number of times that mated of reference vector;
Find out the human body image of the unique and the highest value of the number of times summation that is determined, as the described human body image that the match is successful.
Wherein, be used for calculating the reference vector of absolute value distance of minimum weighting as the highest reference vector of distance.According to the weight of every kind of color, adopt the described distance of absolute value distance formula operation of following weighting, in the robustness that keeps algorithm, reduced time complexity, improved efficiency of algorithm.
Wherein d is the distance of proper vector and reference vector, and x is the proper vector of human body image, the reference vector that X obtains for training, and the figure place of i representation feature vector or reference vector, N is the dimension of proper vector or reference vector, p
iBe every kind of weight that color is corresponding, N is the dimension of proper vector or reference vector.
For aforesaid 6 kinds of colors, for every kind of color is determined a weight, because red wavelength surpasses the wavelength of other color, particularly in the greasy weather, compare with other color, image show more clear.
Through a large amount of test relatively, compare with the identification of weighted not, when the ratio of the weight of redness and other color weight is between 2 ~ 6, can reach 73% ~ 81% accuracy.Particularly the weight of other color is identical, and ratio is 4 o'clock, can reach optimum effect.
For example: if one have 5 human body image samples, each sample is the corresponding a plurality of reference vector of human body image.
The human body image that detects is 5 width of cloth, totally 5 proper vectors; Comprise 5 human body image samples in the database, each sample comprises 6 reference vector, then has 30 reference vector, the corresponding reference vector of everyone volume image.Each proper vector that calculating detects and the distance of 30 reference vector obtain 5 groups of data.
Comprise 30 distances in every group of data, find two minimum distances, d1 and d2, and satisfy 1.5d1＜d2, then think to match reference vector.
Add up the number of times that each reference vector of everyone volume image is mated.For example: detect certain proper vector and be (1,0,0,0,0,0,0,0,0), two nearest reference vector are respectively (0.8,0,0,0,0,0,0,0,0.2) and (0.5,0.5,0,0,0,0,0,0,0) with it.Then can calculate d
1≈ 0.283, d
2≈ 0.707, and 1.5d
1＜d
2Determine that this proper vector and the reference vector that is used for the described d1 of computing are complementary.The reference vector that is used for the described d1 of computing is the human body image of sample 1, and then the human body image of sample 1 is the successful human body image of identification.
Human body image is identified as respectively the situation in following each sample; As: sample 1, sample 1, sample 1, sample 2, sample 3, then statistic histogram is (3,1,1), and sample 1 is high and the most unique sample, and the human body image that is detected finally is identified as sample 1 corresponding human body image again.
In addition, in order to realize exact matching, the human body image that identifies and the image of sample are extracted the ORB unique point, use the hamming distance that unique point is mated, and use RANSAC algorithm eliminating error coupling.Determine whether that according to final matching result the match is successful.
Obviously, those skilled in the art should be understood that, above-mentioned each module of the present invention or each step can realize with general calculation element, they can concentrate on the single calculation element, perhaps be distributed on the network that a plurality of calculation elements form, alternatively, they can be realized with the executable program code of calculation element, thereby, they can be stored in the memory storage and be carried out by calculation element, perhaps they are made into respectively each integrated circuit modules, perhaps a plurality of modules in them or step are made into the single integrated circuit module and realize.Like this, the present invention is not restricted to any specific hardware and software combination.
The above is the preferred embodiments of the present invention only, is not limited to the present invention, and for a person skilled in the art, the present invention can have various modifications and variations.Within the spirit and principles in the present invention all, any modification of doing, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (9)
1. a method of identifying again at the human body in greasy weather is characterized in that, comprising:
Detect the human body image in the video image;
Determine the proper vector of described human body image, according to the weight of different colours, a plurality of reference vector in the proper vector of determining and the database that gathers are in advance mated;
With the human body image that the match is successful in the described database as recognition result.
2. method according to claim 1 is characterized in that, the process of described definite proper vector comprises:
Described human body image is converted to the HSV form;
Add up the interior versicolor pixel quantity of image of described HSV form;
Determine a proper vector corresponding with this image according to described versicolor pixel quantity.
3. method according to claim 2 is characterized in that, the reference vector in the described database is determined by following steps:
Gather in advance everyone several human body images in the video image;
Adopt the K-means clustering algorithm that the proper vector of described several human body images is carried out computing, obtain a proper vector of everyone correspondence as reference vector.
4. method according to claim 3 is characterized in that, described matching process comprises:
The described definite proper vector of computing is the distance of a reference vector corresponding with everyone volume image in the described database respectively;
To a plurality of distance-taxis that each proper vector obtains, determine two minimums apart from d1 and d2; Wherein, d1＜d2;
If described 1.5d1＜d2 determines that then this proper vector and the reference vector that is used for the described d1 of computing are complementary.
5. method according to claim 4 is characterized in that,
Determine the human body image corresponding with the nearest reference vector of each described proper vector, and the summation of the number of times that mated of the reference vector of everyone volume image of statistics correspondence;
Find out the human body image of the unique and the highest value of the number of times summation that is determined, as the described human body image that the match is successful.
6. method according to claim 5 is characterized in that, also comprises: according to the weight of every kind of color, adopt the described distance of absolute value distance formula operation of following weighting;
Wherein d is the distance of proper vector and reference vector, and x is the proper vector of human body image, the reference vector that X obtains for training, and the figure place of i representation feature vector or reference vector, N is the dimension of proper vector or reference vector, p
iBe every kind of weight that color is corresponding, N is the dimension of proper vector or reference vector.
7. method according to claim 6 is characterized in that, red weight and the ratio of other color weight are between 2 ~ 6.
8. method according to claim 6 is characterized in that, also comprises:
The match is successful if do not have, and then the proper vector with described detected human body image joins described database as new reference vector.
9. method according to claim 1 is characterized in that, also comprises:
Current frame image and before video image in, adopt and minimum to state this human body image that detects with color receptacle frame residence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105927217A CN103065128A (en) | 2012-12-30 | 2012-12-30 | Human body reidentification method in foggy days |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105927217A CN103065128A (en) | 2012-12-30 | 2012-12-30 | Human body reidentification method in foggy days |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103065128A true CN103065128A (en) | 2013-04-24 |
Family
ID=48107751
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012105927217A Pending CN103065128A (en) | 2012-12-30 | 2012-12-30 | Human body reidentification method in foggy days |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103065128A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550703A (en) * | 2015-12-09 | 2016-05-04 | 华南理工大学 | Image similarity calculating method suitable for human body re-recognition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030026483A1 (en) * | 2001-02-01 | 2003-02-06 | Pietro Perona | Unsupervised learning of object categories from cluttered images |
| CN101311947A (en) * | 2008-06-12 | 2008-11-26 | 浙江大学 | Real time intelligent control method based on natural video frequency |
| CN102436662A (en) * | 2011-11-29 | 2012-05-02 | 南京信息工程大学 | Human body target tracking method in nonoverlapping vision field multi-camera network |
-
2012
- 2012-12-30 CN CN2012105927217A patent/CN103065128A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030026483A1 (en) * | 2001-02-01 | 2003-02-06 | Pietro Perona | Unsupervised learning of object categories from cluttered images |
| CN101311947A (en) * | 2008-06-12 | 2008-11-26 | 浙江大学 | Real time intelligent control method based on natural video frequency |
| CN102436662A (en) * | 2011-11-29 | 2012-05-02 | 南京信息工程大学 | Human body target tracking method in nonoverlapping vision field multi-camera network |
Non-Patent Citations (1)
| Title |
|---|
| 佟爱华: "一种改进的高精度视频帧间匹配算法", 《仪器仪表学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550703A (en) * | 2015-12-09 | 2016-05-04 | 华南理工大学 | Image similarity calculating method suitable for human body re-recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bedagkar-Gala et al. | Part-based spatio-temporal model for multi-person re-identification | |
| CN103093274B (en) | Method based on the people counting of video | |
| CN105184818B (en) | A kind of video monitoring anomaly detection method and its detecting system | |
| CN103065126A (en) | Re-identification method of different scenes on human body images | |
| CN103049749B (en) | The recognition methods again of human body under grid blocks | |
| CN101470809B (en) | Moving object detection method based on expansion mixed gauss model | |
| CN110490194A (en) | A kind of recognition methods of the multiple features segment fusion traffic sign of adaptive weight | |
| CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
| CN104281853A (en) | Behavior identification method based on 3D convolution neural network | |
| CN104636118A (en) | QR two-dimensional code self-adaptation binarization processing method and device based on light balance | |
| CN103473539A (en) | Gait recognition method and device | |
| CN105825233B (en) | A kind of pedestrian detection method based on on-line study random fern classifier | |
| CN102542244A (en) | Face detection method and system and computer program product | |
| CN102184404B (en) | Method and device for acquiring palm region in palm image | |
| CN103544488B (en) | A kind of face identification method and device | |
| CN110827265A (en) | Image anomaly detection method based on deep learning | |
| CN105184771A (en) | Adaptive moving target detection system and detection method | |
| CN103049747A (en) | Method for re-identifying human body images by utilization skin color | |
| Su et al. | Semantic segmentation of high resolution remote sensing image based on batch-attention mechanism | |
| CN102609715B (en) | Object type identification method combining plurality of interest point testers | |
| CN109240123A (en) | A kind of assemblage on-orbit method and system towards Intelligent logistics vehicle | |
| CN105354570A (en) | Method and system for precisely locating left and right boundaries of license plate | |
| CN103077376B (en) | Method for distinguishing is known again based on the human body image in video image | |
| CN103093273A (en) | Video people counting method based on body weight recognition | |
| CN103065129A (en) | Identifying method for panda |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130424 |
|
| RJ01 | Rejection of invention patent application after publication |